A Brief Introduction to Neural Networks (unfinished series)

This series is intended as a light introduction to neural networks, with a focus on the task of classifying handwritten digits. All programming will be done in python (v3.6).

Support the creation of more tutorials:

[SebastianLague]

Hi everyone, sorry for leaving the series unfinished. I do however have a new video out now which covers everything I intended to talk about in this series (although the code is in C# instead of python). You can watch it here if you’re interested:

[CapnCoCo]

This is by far the best visual example I have ever seen on how the weights and biases change the final output. Great work! Looking forward to the rest of the series.

[jamesallen]

Seriously. I echo everyone else's comments, This is by far the best video I've seen on this topic. Outstanding work.

[XnecromungerX]

I have watched maybe around 50 hours of footage trying to explain topics of neural networks over the years. This is the best one yet.

[runnerpaul]

I'm reading a book which explains ANN's. After four chapters on it I came to the conclusion that I'm not going to get it. Just watched your video by chance and now all I read makes sense. Great job!

[osten]

excuse my french but holy FUCK this is well done. The way you skip unessential bits so that the viewer can grasp the whole before going in to details is great. I personally prefer this mode of education, even when I'm learning seriously, it becomes much faster to learn the details after understanding the goal

[JordanMetroidManiac]

I mean, 3B1B's videos on neural networks are great, and they have some great explanations, but in this video, you explained the "network" part of a neural network in just the right way... in five minutes!!

[JordanMetroidManiac]

Wow... I never knew why the activation function was necessary, but only knew it needed to be there. Now I know! Thanks so much for making this very easy to understand!

[Miszkaaaaa]

After having neural networks at my university - this 5 minute video enlighted me with this visualization. This is pure gold

[chrischeetham]

I have never seen this plotted as a graph before - it's all suddenly very simple! Great video

[AtmosMr]

This is a stunningly clear explanation of how neural networks work, in just under 6 minutes! Amazing. You have exceptional clarity of thought, and it shows in how you explain things. Thank you.

[sinistermephisto]

This is strange i have been taking the machine learning course on Coursera for the past 3 weeks and also checking around online and You by far have the best description of what is going on. You basically get the much faster than anybody has. This is really really strange.
The algebra/calculus and decision boundary. I mean I feel like i should stop the course and wait for you

[wlockuz]

This has to be the most intuitive explanation of a NN, It actually explains why a neural network does what it does in the most simplest terms. Even beats the video by 3b1b if you ask me! Well done!

[adish]

After watching so many videos about neutral networks, I really didn't think that any other way of understanding this thing can exist. But Sebastian, you've done it, again. You have a real talent there...

[thesilversapphire]

As a practitioner, I never had the privilege of being able to explain neural networks with so much ease to audiences who have little or no idea on how activation functions, biases and hidden layers work. You, my friend, are on the top of my list - a big thank you! I look forward to more content like this, keep up the excellent work!

[SriNiVi]

Being a Deep learning researcher and having taken courses both in college and online MOOCs I find this as the best visual explanation of neural networks ever. Huge fan of your channel, been watching it for unity for a while now, and will now start watching your channel for everything. Keep up the great work Sebastian.

[callanbrain]

Best, most intuitive/visual explanation I have seen about how neural networks actually work. Now I actually understand what the activation function is for and I have a better understanding of neural networks in general

[morphman]

It's possibly the best description I've seen so far, but again, I find myself at that point where I ALMOST understand it.

The thing I can't really grasp is what exactly the hidden layer does, or how to determine how many nodes it requires.
I have had some success with some classifications, where I have just randomly changed the number of neurons in the hidden layers, or number of hidden layers, or a combination of both. But without understanding what it does or how to know how many neurons I will need, it's just magic.

And Sebastian here did the same as every other tutor I've come across so far. "And then we'll add a hidden layer with three neurons.." and that's it. That's the explanation we get.

I saw another video, on classification of handwritten numbers, where it was explained that one layer would preferably check for patterns in the writing and the next would check for patterns within those patterns, but no explanation to how that worked when both had the same calculations and the same activation function, nor any explanation on how they came up with that exact number of neurons in each hidden layer.
And when visualizing it, he determined that it didn't check for those patterns at all, yet it did classify the numbers correctly.

So again, without understanding it, it's just down to "magic".

[itameio]

Man, your channel is a treasure. You explain complicated stuff in the easiest and quickest way possible. Best regards.

[iAmTheSquidThing]

This is probably the best explanation I’ve ever seen of a neural network’s inner workings, and in only a few minutes.